Circuit Switching · Computer Networking Interactive

Circuit switching reserves a share of every link before you send anything, which guarantees your rate but wastes whatever you do not use.

Words you will meet

Circuit switching — reserve capacity along the path first, then send.
Circuit — the reserved connection itself.
FDM — frequency-division multiplexing: each circuit gets its own frequency band.
TDM — time-division multiplexing: each circuit gets its own repeating time slot.
Bandwidth — the width of a frequency band, in hertz. Not a speed.
Silent period — time when a connection is reserved but nothing is being sent.
Statistical multiplexing — sharing a link on demand, because users are rarely all busy at once.

Why this matters

This section is the argument the Internet won. Telephone networks reserved capacity for a century, and the designers of the Internet chose not to.

Understanding why explains a great deal of what follows:

why delay is variable, so chapter 3 needs a whole section on estimating it;
why congestion control is needed at all;
why the network makes no promise about the rate you will get.

There are two fundamental approaches to moving data through a network of links and switches. §1.3.1 covered packet switching. Here is the other.

In a circuit-switched network, the resources needed along a path — buffers, link rate — are reserved for the duration of the session. In a packet-switched network they are not: a session’s messages use the resources on demand, and may have to wait in a queue for a link.

Everyday picture — the book’s own analogy

Two restaurants. One requires reservations, the other neither requires nor accepts them.

For the first, you must telephone before leaving home. That is a nuisance. But when you arrive you are seated immediately and can order at once.

For the second, you need not bother reserving. But when you arrive you may have to wait for a table.

Where the picture breaks: a restaurant reservation you do not use is released when you fail to appear, and the table is given to somebody else. A reserved circuit is not. It sits idle for as long as your session lasts, and that is exactly the complaint against circuit switching.

What a circuit actually is

Traditional telephone networks are circuit-switched. Before a sender can send anything, the network must establish a connection between sender and receiver. This is a real connection: the switches on the path maintain connection state for it.

In telephony this connection is called a circuit . When the network establishes it, it also reserves a constant rate on each link along the way, for the whole duration of the connection. Because that rate is held aside, the sender can transfer data at a guaranteed constant rate.

Figure 1.13 — four switches, four links, four circuits on each link

4 hops: Host A → switch 1 → switch 2 → switch 4 → Host B

Drag any device to rearrange the picture. Hover a link to see its rate, delay and length.

Each link carries four circuits, so each link supports four simultaneous connections. To connect Host A to Host B the network reserves one circuit on each of two links — the book uses the second circuit on the first link and the fourth on the second. With 1 Mbps links, that connection gets a guaranteed 250 kbps for as long as it lasts.

Read this diagram as text

Host A — wired link — switch 1
Host C — wired link — switch 2
Host B — wired link — switch 4
switch 1 — wired link — switch 2
switch 2 — wired link — switch 4
switch 3 — wired link — switch 4
switch 1 — wired link — switch 3

Work through the numbers on that diagram. Each link carries four circuits, so each link supports four simultaneous connections. If each link runs at 1 Mbps, then each end-to-end connection receives one quarter of it:

\text{rate per circuit} = \frac{1\ \text{Mbps}}{4} = 250\ \text{kbps}

Contrast that with a packet-switched network. There the packet is sent with no reservation at all. If a link is congested because other packets need it at the same time, the packet waits in a buffer and suffers a delay. The Internet makes its best effort to deliver packets in a timely manner, but it makes no guarantees.

How a link is carved into circuits

A circuit inside a link is implemented in one of two ways.

With FDM (Frequency-Division Multiplexing) , the frequency spectrum of the link is divided among the connections. The link dedicates a frequency band to each connection for the whole connection. In telephone networks that band is typically 4 kHz. The width of the band is called the bandwidth .

FM radio stations use FDM too, sharing the spectrum between 88 MHz and 108 MHz with each station allocated its own band.

With TDM (Time-Division Multiplexing) , time is divided into frames of fixed duration, and each frame is divided into a fixed number of time slots. When the network establishes a connection, it dedicates one time slot in every frame to that connection.

Figure 1.14 — the same link, divided two different ways

following circuit 2 of 4

FDM: circuit 2 owns 4 kHz of the spectrum and uses it all the time — a thin slice, always.

TDM: circuit 2 owns the whole link — but only during 1 slot in every 4. At 8,000 frames/s and 8 bits per slot that is 64 kbps.

Press play. Under FDM circuit 2 is lit the whole time on its own narrow band. Under TDM it is lit only one slot in four, but across the entire link. Same average share, opposite rhythm.

In plain words

Both methods give a circuit the same average share of the link. They differ in how that share arrives.

FDM hands you a narrow lane you may drive in continuously.
TDM hands you every lane, but only for one second in four.

Neither is faster. The picture above shows the same four circuits both ways.

The rate of a TDM circuit

For TDM, the rate of a circuit is the frame rate multiplied by the number of bits in a slot:

R_{\text{circuit}} = (\text{frames per second}) \times (\text{bits per slot})

frames per second — how often the pattern of slots repeats (hertz).
bits per slot — how much data fits in this circuit’s slot (bits).

Read aloud: your rate is how much you send each turn, times how often your turn comes round.

Worked example — the book’s own numbers

A link transmits 8,000 frames per second, and each slot holds 8 bits.

R_{\text{circuit}} = 8{,}000\ \text{frames/s} \times 8\ \text{bits} = 64{,}000\ \text{bits/s} = 64\ \text{kbps}

That 64 kbps is not an arbitrary example. It is the rate of one digital telephone channel, and it is why so many telephone-network numbers are multiples of 64.

The worked example: sending a file over a circuit

Live calculator — the book’s worked example

What each symbol means

F — the size of the file to send (bits)
link rate — the rate of each link, shared by all its circuits (bits per second)
slots — how many circuits the link is divided into (a count)
setup — the time to establish the circuit before any data flows (milliseconds)

Read aloud: a circuit gets the link rate divided by the number of slots; the file takes its size divided by that rate, plus the time to set the circuit up.

File size FbitsLink rateMbpsTDM slots per framecircuitsCircuit setup timems

Total time to send the file10.5 s

rate of one circuit = link rate / slots
                    = 1.54 Mbps / 24 = 64.0 kbps
transmission time   = F / rate
                    = 640,000 bits / 64.0 kbps = 10.0 s
total               = 10.0 s + 500 ms setup
                    = 10.5 s
(the number of links on the path does not appear anywhere above)

Change any number above and the arithmetic re-runs, carrying the units through.

The book’s answer is 10.5 seconds. Notice what is missing from the inputs: the number of links on the path. It genuinely does not matter.

The book’s setup: send a file of 640,000 bits across a circuit-switched network. Every link uses TDM with 24 slots and runs at 1.536 Mbps. It takes 500 ms to establish the circuit before Host A can begin.

Each circuit therefore gets 1.536 Mbps ÷ 24 = 64 kbps. The file takes 640,000 ÷ 64,000 = 10 seconds, and adding the setup gives 10.5 seconds.

The detail worth pausing on

The transmission time does not depend on the number of links. It would be 10 seconds whether the circuit passed through one link or a hundred.

Compare that with equation 1.1, where a packet costs a fresh $L/R$ at every single link. This is what you buy with the setup time and the reservation: the path stops mattering.

Why packet switching won

Critics of packet switching argue that it is unsuitable for real-time services such as telephone and video calls. Their reason is that its end-to-end delays are variable and unpredictable, mostly because of queuing.

Proponents answer with two claims: packet switching shares capacity better, and it is simpler, more efficient and less costly to build. Both are worth examining.

Reserved capacity goes to waste

The dedicated circuits are idle during silent periods. When one person in a telephone call stops talking, the reserved frequency band or time slot cannot be used by any other connection. It simply goes unused.

The book’s sharper example is a radiologist using a circuit-switched network to look at x-rays remotely. The radiologist establishes a connection, requests an image, contemplates it, then requests another. Throughout the contemplation the network resources are allocated to that connection and doing nothing at all.

Bursty users fit better than reserved ones

Live calculator — why 35 users fit where only 10 were reserved

What each symbol means

link rate — the capacity of the shared link (bits per second)
per user — what one user needs while actually sending (bits per second)
M — how many users share the link under packet switching (a count)
p — the fraction of time each user is active (a probability)
n — how many can be active at once before the link overflows (a count)

Read aloud: circuit switching must reserve for every user at once; packet switching only has to cope with the number that are actually busy together.

Shared link rateMbpsEach active user needskbpsUsers on the link (packet switching)usersEach user is activeof the timeLink copes with this many at onceusers

Circuit switching supports10 users

users = link rate / rate reserved per user
      = 1.00 Mbps / 100 kbps
      = 10 users, reserved whether or not they are sending

Packet switching: chance of overload0.042 %

P(more than 10 of 35 users active at once), each active with p = 0.1
P(at most 10 active) = 0.999576
P(more than 10)      = 1 − 0.999576 = 4.243e-4
so the link is overloaded only about 4.2 times in 10,000 moments

Change any number above and the arithmetic re-runs, carrying the units through.

The book’s numbers: 10 users under circuit switching, and with 35 users under packet switching the chance that more than 10 are active at once is about 0.0004. Three and a half times the users, at essentially the same quality.

Suppose users share a 1 Mbps link. Each user alternates between activity, generating data at 100 kbps, and inactivity, generating nothing. Each user is active only 10 % of the time.

With circuit switching, 100 kbps must be reserved for every user at all times, so the link supports 10 users and no more.

With packet switching, put 35 users on it. The chance that 11 or more are active simultaneously is about 0.0004. For the other 99.96 % of the time the total arriving is at most 1 Mbps, so packets flow essentially without delay — just as they would under circuit switching.

More than three times the users, at essentially the same quality. This on-demand sharing is called statistical multiplexing .

One busy user fits better too

The second example is smaller and even more pointed. Ten users share the link. One of them suddenly has a thousand 1,000-bit packets to send. The other nine are silent.

Under TDM circuit switching with 10 slots per frame and 1,000 bits per slot, the active user may use only its own slot. The remaining nine slots in every frame stay empty. Sending one million bits takes 10 seconds.

Under packet switching, the active user sends continuously at the full 1 Mbps, because nobody else has anything to multiplex with. The same data takes 1 second.

In plain words — the crucial difference

Circuit switching pre-allocates the link regardless of demand, and allocated but unneeded time goes unused.

Packet switching allocates on demand, so the link is shared packet by packet among only those users who actually have something to send.

The two approaches together

Circuit switching and packet switching, side by side

	Circuit switchingreserve first	Packet switchingsend on demand
Before any data moves
Rate the sender gets
End-to-end delay
When the sender goes quiet
Users supported on a 1 Mbps link
Implementation
Where it is used

Cells marked ⓘ have a reason behind them — click to read it.

Both are still in use, but the trend has clearly been toward packet switching. Even today’s circuit-switched telephone networks are slowly migrating — they often use packet switching for the expensive overseas portion of a call.

Check yourself

Check yourself — §1.3.2

0 of 6 answered

1.What exactly does a circuit-switched network reserve?
2.predictPress play with the diagram showing both FDM and TDM. What is different about how circuit 2 sends?
Watch the FDM band and the TDM strip at the same time.
3.predictIn the file-transfer calculator, change the number of links the circuit crosses. What happens to the transmission time?
Look for the number of links among the inputs.
4.A 1 Mbps link serves users who each need 100 kbps when active, and who are active 10 % of the time. Why can packet switching support 35 of them when circuit switching supports 10?
5.One user has a thousand 1,000-bit packets to send and the other nine are idle. TDM circuit switching takes 10 seconds; packet switching takes 1. Why?
6.What is the strongest argument the book gives *against* packet switching?

What to remember

Circuit switching reserves buffers and link rate along the whole path before any data moves. Packet switching reserves nothing.
A link is divided into circuits by FDM (a band each, all the time) or TDM (a slot each, in turn). Same share, different rhythm.
Because users are bursty, statistical multiplexing lets 35 users share a link sized for 10, with about a 0.0004 chance of overload. That is why packet switching won.

Two ways to share a link